ICRM3: 3rd International Chemometrics Research Meeting, Veldhoven, The Netherlands, May 26–30, 2002
Henk‐Jan Ramaker, Uwe Thissen · Journal of Chemometrics · 2003
The aim of the ICRM meetings is to provide chemometricians worldwide the opportunity to listen to a well-balanced mix of high quality presentations.This mix exists of extensions of existing research but also of new research.However, the most appealing approach of the ICRM is the fact that every day only two lectures are given (each of an hour), followed by a discussant reflecting on the lectures (10 minutes).After that sufficient time is available for the audience to participate in the discussion (20 minutes).The topics of the lectures ranged from process chemometrics, multivariate calibration and classification, multiway and multiblock issues, data mining, combining hard and soft models, nonlinear multivariate analysis, and spectral imaging.Additionally, a total of 58 posters have been presented together with a commercial exhibition.The meeting was started by Joe Qin (University of Texas, Austin TX, USA) who introduced the definitions of process chemometrics.In case of multivariate statistical process control, faults are detected in the SPE or D-chart.A next step might be to perform fault identification.It is explained how fault identification by construction can be performed.All possible (multivariate) fault-directions are constructed from historical data.Then the actual fault is compared to these fault directions.Paul Gemperline (East Carolina University, Greenville NC, USA) presented a well-organised introduction to a method which combines spectroscopic and calorimetric measurements to characterise batch reactions using self-modelling curve resolution and kinetic fitting.The goal was to combine the data into a model of spectral profiles, reaction rates, mass balance, and energy balance.This model then can be used for the optimisation of batch recipes.Harald Martens (Royal Veterinary and Agricultural University, Copenhagen, Denmark) started the presentation by giving a nice overview of 5 methods to combine new observations with old knowledge: PLS regression, Bookstein's PLS, bridge PLS regression, L-PLSR and G-PLS.The first three methods were clarified with some experimental data of Harald's brains.It turned out these brains work rather randomly.Furthermore, Harald showed that in some cases EMSCpreprocessing of spectra can be helpful.Lars Nørgaard (Royal Veterinary and Agricultural University, Copenhagen, Denmark) started his talk rather provoking by wearing a Danish soccer hat (before the match against England).Various topics were discussed about calibration and classification.First, a data screening approach was suggested by using interval PLS and PCA.Then, three different classification methods were discussed: i) classifying in PCA with Mahalanobis distance ii) CVA and iii) an